Why convert a PDF to Markdown instead of plain text for AI and notes?
Markdown retains document hierarchy, bullet lists, and data tables using lightweight ASCII punctuation rather than stripping everything into flat paragraphs. Exporting to the standardized IETF RFC 7763 text/markdown Media Type gives AI chat tools, knowledge bases, and static site generators clear structural cues without uploading private PDFs to cloud parsers.
Preserving semantic headings and lists for LLMs
When you paste raw text from a PDF into ChatGPT or Claude, section titles blend into body paragraphs and multi-column tables scramble across lines. Markdown marks major titles with # and ##, which helps chunkers and attention windows keep related paragraphs grouped together.
For raw prose where you do not want hash marks or pipe delimiters at all, you can switch to plain .txt text extraction. Otherwise, Markdown delivers much higher signal per token on structured reports, specs, and contracts.
Clean ingestion into Notion, Obsidian, and GitHub
Notes apps like Obsidian and Notion, as well as repository README files, parse .md natively into styled headings, checklists, and tables. Because the conversion finishes inside your browser tab with 0 bytes transmitted, confidential meeting packets and internal SOPs stay private until you choose where to paste them.
Live sidebar counters for Words, Headings, and Tables update immediately as you toggle options. Watching those numbers confirms right away that your document’s sections and tabular blocks were recognized before you copy.
Embedded pictures and vector drawings are intentionally omitted during conversion so the Markdown output remains clean, portable text without binary image blobs or broken local file paths.
How heading levels, bullet lists, and pipe tables are determined
The converter analyzes positioned text lines from pdf.js rather than treating the page as an unstructured character stream. Measuring font sizes, vertical spacing, and horizontal column gaps lets the formatter rebuild structure deterministically.
Font-size heading detection and CommonMark escaping
First, the parser calculates the median body font size across the entire document, falling back to 12 pt if needed, so heading levels remain consistent from page 1 to the end. Any line exceeding 1.15 times the median body size promotes to an ATX heading (# or ##).
Normal prose lines that happen to start with #, >, +, -, or 1. are automatically backslash-escaped so regular sentences do not turn into accidental blockquotes or lists. Short bold lines under 200 characters wrap in **...** for emphasis, while lettered list items keep their original prefix.
Column-gap table detection and merged-cell limits
To build pipe tables matching the GitHub Flavored Markdown (GFM) Tables Specification, the layout engine scans for at least 3 consecutive lines that split cleanly at wide horizontal gaps into 2 or more cells. Those runs become tables when they span at least 3 columns or average under 28 characters per cell, preventing two-column magazine prose from being misread as a table.
Complex spreadsheets with merged header spans or multi-line wrapped cells can confuse coordinate-based column splitting. If a table looks misaligned, unticking Turn tables into Markdown tables falls back to clean sequential lines. If you are starting from a screenshot of a table rather than a PDF, you can convert screenshots into structured Markdown using the same layout engine.
Toggle options for tables, repeated headers, and page markers
Three checkboxes in the options panel re-run the Markdown formatter in memory instantaneously without re-reading the PDF, outputting syntax compatible with the CommonMark 0.31.2 Specification. Toggling them lets you tailor the output for AI prompts, archival notes, or human editing.
| Sidebar option | Default | What it changes in output | When to toggle it |
|---|---|---|---|
| Turn tables into Markdown tables | On | Formats aligned columns as pipe tables | Untick if merged cells scramble rows |
| Leave out page headers, footers and numbers | On | Strips repeating top/bottom edge lines | Untick if every line is short content |
| Mark where each page starts | Off | Inserts <!-- Page N --> comments |
Tick when citing exact PDF page numbers |
| Automatic bold and list detection | Always on | Preserves short bold lines and list markers | Built in for clean document hierarchy |
When Leave out page headers, footers and numbers is enabled on PDFs with 3 or more pages, the filter checks the outer 10% of pages whose vertical span reaches at least 180 pt. Any running header or footer pattern appearing on at least half the pages, along with standalone page numbers, is removed so paragraphs flow uninterrupted across page breaks.
If you have a 60-page report and want key takeaways before reading every section, you can also generate a quick summary of long PDFs right in your browser.
Local OCR on scanned PDFs and exporting clean .md files
If a PDF contains scanned books, articles, or paper memos with no selectable text layer, the tool automatically invokes self-hosted Tesseract OCR on those pages. Recognized words return with bounding boxes and font-size estimates, so scanned pages still produce headings, lists, and pipe tables.
Tip: If your PDF consists solely of repeated headers or short labels and the preview warns that every line was filtered out, untick Leave out page headers, footers and numbers to restore all lines immediately.
When the Markdown looks right, click Copy Markdown to copy it to your clipboard (with a 1,600 ms Copied confirmation) or click Save .md to download a UTF-8 file. Later, if you revise the Markdown notes and want a formatted handout, you can render edited Markdown notes back into PDF locally.
Because the OCR engine runs in your browser tab without server queues, processing long scanned books takes a few seconds per page. Once finished, all three formatting toggles remain interactive so you can refine tables or page markers before exporting.